With the advent of the era of big data and artificial intelligence, more and more fields are choosing emerging technological methods to improve work efficiency. Whether in the fields of production, technology, or education, among others, there is a gradual shift towards using artificial intelligence to replace traditional work methods. Analyzing the behavioral status of university students in the teaching and management process is an effective means of improving teaching level and academic performance. Traditional methods of studying classroom behavior generally require teachers to concentrate and watch replays of classroom videos. As a big data analysis technology, using deep learning to study student classroom behavior can alleviate the pressure on teachers, allowing them to devote more energy to teaching, improve teaching methods, and guide students to better engage in learning in the classroom. In order to reduce the requirements for deploying devices in the classroom scene and improve recognition accuracy and speed, this paper proposes a deep learning-based model for recognizing student classroom behavior, building upon existing recognition models. Firstly, a fusion ghost module is introduced to achieve lightweight models, and then a coordinate attention mechanism is incorporated to enhance detection accuracy. Experimental results show that the improved model achieves higher recognition accuracy compared to the original model, while reducing inference time and parameter count.